(upbeat music)
(gears whirring) - Coming up, we take a look
at how anyone can easily build automated, no code, chatbot experiences with power virtual agents,
which gives you a fast, intuitive, and AI-rich way
to create intelligent bots that make decisions and take actions. So, we're joined today by Emma Archer from the Power Virtual Agents team. Welcome to Microsoft Mechanics. - Thank you for inviting me. - Thanks again for joining us today, Emma. Now, before we get started and go further, for those of you who are
new to Power Virtual Agents, it's Microsoft's newest
member of the Power Platform, our family of tools for low to zero code modern app
development, process automation, business insights, and now, chatbots. Now Power Virtual Agents: this isn't your average
chatbot creation service. But what's different about this approach? - A few things, Power
Virtual Agents is a SaaS, which means there's no set up involved, you just sign up and start
building your chatbot. And we're not talking about
building a simple Q&A chatbot, it goes way beyond that. Quite literally, anyone
can build a chatbot, our graphical user interface
makes it super easy to get started. You can seed your chatbot with FAQ content from existing websites, author
topics with natural language using entities, variables,
and slot-filling to jump to pertinent questions. The chatbot can take actions
based on specific conditions using Power Automate and connectors to connect to your backend systems. And you can extend your chatbot with Azure bot framework skills. You can also publish, with a single click, to different channels. And you can monitor the
chatbot's performance and continuously make
improvements over time. - And what I really
like about this approach is that it really puts
the power in the hands of subject matter experts. They don't have to be able
to code, or be a developer, or have data science skills. But can we see one of these in action? - Absolutely. So, let me start by showing you
the end customer experience, and then I'll show you how to build one. I have here a chatbot on
the Contoso flooring site, and it's designed to help
customers get a cost estimate for a new floor. So let's go ahead and do that. I'm actually interested
in hardwood flooring. And I am a returning customer, so let me pop in my email address so it can go ahead and look up my details. And you can see it welcomes me back and asks me to confirm that home address. It's correct. And now it's offering me the
next available time slot. And that works for me, so
I'll go ahead and take it. And everything's scheduled. I now have an appointment with Beth and I'll get an email confirmation. - And this is a really
natural set of interaction, and kind of a natural
experience and conversation. Now the chatbot isn't just conversing, it's also taking actions too. So there's a lot here
going on behind the scenes, but how did you build it? - The build experience
is really the best part. It's super easy to get started. You just go to
powervirtualagents.microsoft.com and click on the start free. And once you've signed up you're going to land in the
Power Virtual Agent's designer, and you're ready to
create your first chatbot. You simply give it a
name and click Create. Now, I've already created a chatbot so I can go ahead and get
a headstart on the design by ingesting content from
an existing FAQ site. So I'll just copy the
URL, come into Topics, Suggested, Get Started, paste
that URL and click Start. Now, it'll take a minute or
two to ingest the content. So, you can see here that the
AI has successfully extracted the content and created
suggestions to review. Now, I can either choose
to promote en masse, or more likely I'm going
to review individually and determine which ones need
a tweak before I accept them. - So how did you enable then the natural language
conversation aspect of that? - We're using natural
language understanding. Power Virtual Agents is
built on years of AI research and applied science. We have a universal language model that we've trained with
a huge amount of data. It's been optimized for the
way that people use language when interacting with chatbots. And that enables us to
focus on what you mean rather than what you type. You can see here we have
several trigger phrases for our flooring estimate topic,
and it's a good idea to use between five and 10 to train
the natural language model. I'll type in a phrase which is somewhat semantically similar. I want to get a kitchen flooring quote. And you can see that, even
though my wording doesn't match any of the trigger phrases,
the chatbot understood and successfully launched
the correct topic. - So it's pretty easy,
then, to get started. When you showed us the chatbot, the experience on the website itself, the customer's actually
specifying a specific floor type. But how is it able to recognize really what the customer was
requesting in this case? - Well, our natural language
understanding helps with that. We have what we call an entity, and that's basically a
category of information such as their phone number, city, zip, or even a person's name. And Power Virtual Agents comes with a number of
predefined entities, and you have the ability to
define your own custom entities, which is what I've done
here with flooring types. You can see that I've
specified the various options that are available to get a quote about. And I've added a couple
of synonyms for hardwood. So, how is this used, this custom entity,
within the topic itself? Within our flooring estimate,
in the question node you can select the custom
flooring type entity from that dropdown list. And you can also have the
option to define what buttons are available to the customer. For example, if I perhaps
don't want to show carpet as an option, I can deselect it. You can also see here a
couple of conditional branches for hardwood and laminate,
but I have nothing for tile. So I'll go ahead and add a
conditional node for that and it will create a new
branch for me to populate. - And all of this kind of
helps make that bot interaction more natural, right? - It does, and now I've added my entity to the topic dialogue
tree, there's more I can do with a capability called slot filling. This gives the chatbot
even more flexibility in how it interacts with you. For example, if the
first thing you asked for is a quote for hardwood flooring, it can skip over that question
about what type of flooring and go directly on to the next one, making the interaction appear more human. - So this makes the process
a lot more efficient and the interaction itself
is more intelligent. And it was really able to
see and also take actions, and it was able to set up,
in this case, an appointment with a flooring specialist,
but how did that all work? - Great question. This is where Power Automate comes in. It makes it super easy to perform actions. You can see we're in the
flooring topic right now, and because we have track
between topics switched on, you'll be able to see our
progress across the topics as we continue our conversation
here in the test window. I click hardwood and it asks
me if I'm a returning customer, and I say yes, and, at this point, it's going to navigate us
over to the other topic, schedule an appointment
for returning customer. And this topic is calling
two Power Automate flows. The first takes the email
address that I provide and does a lookup to
the Dynamic 365 instance to find the customer information
and confirm it back to me. And the second is going to look up the next available time slot for an agent to do a home visit. And you can see here, this is
the schedule appointment flow, and we can deep link into Power Automate to make edits as needed. - Okay, so now you've
got everything built, but how do I go about then
publishing the chatbot? - Once you're done with all the authoring and testing of your chatbot, you'll want to deploy it to
the appropriate channels. And you can see here
the available choices, the most common being custom website. And you can navigate to Publish,
click on the Publish button to deploy the latest
version of your content to any of the channels that
you've already got configured. - Okay, so you've basically written once and published anywhere
using supported connectors, but now you mentioned before you can also look at the metrics that are kind of behind
the chatbot experience. - That's right. So, with Power Virtual Agents you can monitor the
performance of your chatbot and determine what changes you
need to make to improve it. Probably one of the most important things. On our analytic summary page here you can see the total sessions
for a selected time period. A session basically starts when
an end user starts chatting with a bot, and it ends when
a problem is either resolved, escalated to a human agent, or abandoned. The engagement rate is a breadth measure, and it shows you what
portion of the sessions the chatbot had content for. I.E. was able to trigger a topic for. To review the unengaged sessions, you can go to the sessions tab and download the chat transcripts to review what topics
potentially you need to add. The next three measures
show you what portion of engaged sessions were
resolved successfully, versus escalated to a human
agent, versus abandoned. And, obviously, the
higher the resolution rate the better performing your bot. You'll want to focus in on the topics that have a high escalation
rate and a high abandoned rate. And review the chat
transcripts to determine what improvements you can
make to those specific topics. - This is really good to see
really how much easier it is now to build these types
of intelligent chatbots from scratch, but what's next
for you then and the team? - So we're adding a multi-user capability so you can bring in
numerous content experts to author the chatbot at the same time. And we're also adding the
ability to create the chatbot in one of 20-plus languages. And a lot more in the coming months. - Thanks again so much for
joining us today, Emma, and really that introduction
to Power Virtual Agents. Now, if you're interested in
testing this out for yourself, you can go to
powervirtualagents.microsoft.com. Thanks for watching, everybody, and we'll see you next
time, goodbye for now. (upbeat music)